Hi, from what I've understood about the nature of the project, before loading new
work, the precedent batch needs to be completed. This naturally induces people to store a lot of WUs in their cache, which in turn causes a general slow down.
Has it ever been thought to limit the number of WUs a device can store? Or even completely eliminate the possibility to do it? This could significantly decrease the time needed to complete a batch, and everyone would be able to get work until the batch is *actually* finished.

Hi, from what I've understood about the nature of the project, before loading new
work, the precedent batch needs to be completed. This naturally induces people to store a lot of WUs in their cache, which in turn causes a general slow down.
Has it ever been thought to limit the number of WUs a device can store? Or even completely eliminate the possibility to do it? This could significantly decrease the time needed to complete a batch, and everyone would be able to get work until the batch is *actually* finished.

Its limited to 50 per core depends on the runtime. Yes I agree that the turnaround time is a bit low at the moment.

welcome David!